Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Feedback Context Reweights Learning Mechanisms during Auditory Category Learning
Poster Session A, Wednesday, September 30, 11:00 am - 1:00 pm, Wangari Maathai
Zhenjiang CUI1, Zhenzhong GAN3, Suiping WANG3, Gangyi FENG1,2; 1Department of Linguistics and Modern Languages, The Chinese University of Hong Kong, 2Brain and Mind Institute, The Chinese University of Hong Kong, 3Philosophy and Social Science Laboratory of Reading and Development in Children and Adolescents, South China Normal University
Learning speech requires forming stable categories from highly variable acoustic input. This is especially challenging in adulthood, when new phonemes or lexical tones must be learned through experience, feedback, and communicative outcomes. Although feedback is known to shape speech and auditory category learning, it remains unclear whether different feedback contexts merely change learning success or instead alter the relative engagement of multiple learning mechanisms. Here, we tested whether feedback content reweights reinforcement-based, statistical, and associative learning mechanisms during auditory category learning, and whether these computational differences are reflected in neural representational geometry. Here, we tested this question using a controlled auditory category-learning paradigm designed to isolate mechanisms relevant to speech learning. Ninety-three participants learned to categorize 40 novel ripple sounds across six training blocks during fMRI. They were assigned to full-feedback, minimal-feedback, or no-feedback conditions. Full feedback provided both correctness and the category label; minimal feedback provided correctness only; and no feedback provided neither. Behaviourally, both feedback groups showed robust learning (last vs. first block: ACC/RT, t = 5.38; Ps < 0.001), whereas the no-feedback group showed substantially weaker learning improvement (last vs. first block: ACC/RT, t = 2.63; P = 0.014). To characterize the underlying computations, we fit trial-by-trial responses using reinforcement, statistical, and associative learning models. Model comparison showed distinct profiles across feedback conditions. Full-feedback behavior was best explained by a reinforcement-learning model incorporating correct-response reinforcement and explicit category label, outperforming all others (all ΔBIC > 5). Minimal-feedback behavior was best explained by a statistical-learning model tracking distributional properties without category labels (ΔBIC > 18). No-feedback behavior was best fit by an associative-learning model without feedback updating, surpassing a random model (ΔBIC = 65.12). These findings indicate that feedback context shifts the reliance on different learning mechanisms. Model-derived representational dissimilarity matrices were then used to examine acoustic-, perceptual-, and decision-level neural representations. Full feedback strengthened decision-level representations in sensorimotor cortex and perceptual-level representations in orbitofrontal, parietal, and posterior inferior temporal regions. Minimal feedback primarily enhanced perceptual-level structure across frontal, parietal, and temporal cortices, with weaker decision-level involvement. Without feedback, perceptual representations were limited, whereas decision-related activity was relatively preserved, consistent with greater reliance on stimulus-response association. These findings show feedback enhances speech learning by altering information, reweighting neural mechanisms, and supporting category formation. Rich feedback biases learning toward reinforcement and perceptual representations, minimal feedback emphasizes statistical structure, and no feedback relies on stimulus-response associations. Adult speech category learning success depends on a flexible balance among multiple mechanisms.
Topic Areas: Speech Perception, Computational Approaches